[Mesa-dev] [PATCH v4 1/3] nir: Add a pass to lower vector phi nodes to scalar phi nodes

[Ian Romanick]

On 01/27/2015 05:31 PM, Jason Ekstrand wrote:
> v2 Jason Ekstrand <jason.ekstrand at intel.com>:
>  - Add better comments
>  - Use nir_ssa_dest_init and nir_src_for_ssa more places
>  - Fix some void * casts
> 
> v3 Jason Ekstrand <jason.ekstrand at intel.com>:
>  - Rework the way we determine whether or not to sccalarize a phi node to
>    make the recursion non-bogus
>  - Treat load_const instructions as scalarizable
> ---
>  src/glsl/Makefile.sources               |   1 +
>  src/glsl/nir/nir.h                      |   2 +
>  src/glsl/nir/nir_lower_phis_to_scalar.c | 266 ++++++++++++++++++++++++++++++++
>  3 files changed, 269 insertions(+)
>  create mode 100644 src/glsl/nir/nir_lower_phis_to_scalar.c
> 
> diff --git a/src/glsl/Makefile.sources b/src/glsl/Makefile.sources
> index face22e..bf6b70b 100644
> --- a/src/glsl/Makefile.sources
> +++ b/src/glsl/Makefile.sources
> @@ -31,6 +31,7 @@ NIR_FILES = \
>  	nir/nir_lower_global_vars_to_local.c \
>  	nir/nir_lower_locals_to_regs.c \
>  	nir/nir_lower_io.c \
> +	nir/nir_lower_phis_to_scalar.c \
>  	nir/nir_lower_samplers.cpp \
>  	nir/nir_lower_system_values.c \
>  	nir/nir_lower_to_source_mods.c \
> diff --git a/src/glsl/nir/nir.h b/src/glsl/nir/nir.h
> index 980fdd0..4f58eee 100644
> --- a/src/glsl/nir/nir.h
> +++ b/src/glsl/nir/nir.h
> @@ -1526,6 +1526,8 @@ void nir_remove_dead_variables(nir_shader *shader);
>  void nir_lower_vec_to_movs(nir_shader *shader);
>  void nir_lower_alu_to_scalar(nir_shader *shader);
>  
> +void nir_lower_phis_to_scalar(nir_shader *shader);
> +
>  void nir_lower_samplers(nir_shader *shader,
>                          struct gl_shader_program *shader_program,
>                          struct gl_program *prog);
> diff --git a/src/glsl/nir/nir_lower_phis_to_scalar.c b/src/glsl/nir/nir_lower_phis_to_scalar.c
> new file mode 100644
> index 0000000..a94b8b0
> --- /dev/null
> +++ b/src/glsl/nir/nir_lower_phis_to_scalar.c
> @@ -0,0 +1,266 @@
> +/*
> + * Copyright © 2015 Intel Corporation
> + *
> + * Permission is hereby granted, free of charge, to any person obtaining a
> + * copy of this software and associated documentation files (the "Software"),
> + * to deal in the Software without restriction, including without limitation
> + * the rights to use, copy, modify, merge, publish, distribute, sublicense,
> + * and/or sell copies of the Software, and to permit persons to whom the
> + * Software is furnished to do so, subject to the following conditions:
> + *
> + * The above copyright notice and this permission notice (including the next
> + * paragraph) shall be included in all copies or substantial portions of the
> + * Software.
> + *
> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
> + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
> + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
> + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
> + * IN THE SOFTWARE.
> + *
> + * Authors:
> + *    Jason Ekstrand (jason at jlekstrand.net)
> + *
> + */
> +
> +#include "nir.h"
> +
> +/*
> + * Implements a pass that lowers vector phi nodes to scalar phi nodes when
> + * we don't think it will hurt anything.
> + */
> +
> +struct lower_phis_to_scalar_state {
> +   void *mem_ctx;
> +   void *dead_ctx;
> +
> +   /* Hash table marking which phi nodes are scalarizable.  The key is
> +    * pointers to phi instructions and the entry is either NULL for not
> +    * scalarizable or non-null for scalarizable.
> +    */
> +   struct hash_table *phi_table;
> +};
> +
> +static bool
> +should_lower_phi(nir_phi_instr *phi, struct lower_phis_to_scalar_state *state);
> +
> +static bool
> +is_phi_src_scalarizable(nir_phi_src *src,
> +                        struct lower_phis_to_scalar_state *state)
> +{
> +   /* Don't know what to do with non-ssa sources */
> +   if (!src->src.is_ssa)
> +      return false;
> +
> +   nir_instr *src_instr = src->src.ssa->parent_instr;
> +   switch (src_instr->type) {
> +   case nir_instr_type_alu: {
> +      nir_alu_instr *src_alu = nir_instr_as_alu(src_instr);
> +
> +      /* ALU operations with output_size == 0 should be scalarized.  We
> +       * will also see a bunch of vecN operations from scalarizing ALU
> +       * operations and, since they can easily be copy-propagated, they
> +       * are ok too.
> +       */
> +      return nir_op_infos[src_alu->op].output_size == 0 ||
> +             src_alu->op != nir_op_vec2 ||
> +             src_alu->op != nir_op_vec3 ||
> +             src_alu->op != nir_op_vec4;

I don't think this logic is correct.  It will return true if src_alu->op
is nir_op_vec2 because that's not nir_op_vec4.  Right?